Some projects I did while learning about generative adversarial networks from online lectures of Ian Goodfellow (DCGAN) and Jun Yan Zhu (CycleGAN). I also wrote more detailed explanation of project 3 and 4 in their folders. I find the concept of having advarsarial networks competing to minimize their individual losses with the expense of the other to be incredibly innovative. Also, using the concept of Cycle-Consistency-Loss to incorporate an element of supervised learning for producing images out of thin air is interesting and fun to implement as well. I will continue looking into GAN and find out just how dangerous it becomes.
- Generative Adversarial Networks
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
- Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
- Least Square Generative Advarsarial Networks
- Deep Residual Learning for Image Recognition